import gradio as gr
from llama_cpp import Llama
llm_f16 = Llama(model_path="./qwen-1.8b-f16.gguf",
            n_ctx=4096,
            n_threads=2,
            chat_format="chatml")
llm_q5_k_m = Llama(model_path="./qwen-1.8b-q5_k_m.gguf",
            n_ctx=4096,
            n_threads=2,
            chat_format="chatml")

def chat_stream_completion(message, history, system_prompt, q5_check):
    messages_prompts = [{"role": "system", "content": system_prompt}]
    llm = None
    if q5_check:
        llm = llm_q5_k_m
    else:
        llm = llm_f16

    for human, assistant in history:
        messages_prompts.append({"role": "user", "content": human})
        messages_prompts.append({"role": "assistant", "content": assistant})
    messages_prompts.append({"role": "user", "content": message})
    response = llm.create_chat_completion(
        messages=messages_prompts,
        stream=True,
        stop="\n\n\n"
    )
    message_repl = ""
    for chunk in response:
        if len(chunk['choices'][0]["delta"]) != 0 and "content" in chunk['choices'][0]["delta"]:
            message_repl = message_repl + \
                chunk['choices'][0]["delta"]["content"]
        yield message_repl

gr.ChatInterface(
    chat_stream_completion,
    additional_inputs=[gr.Textbox(
        "You are helpful AI.", label="System Prompt"), gr.Checkbox(label="Use Q5-K-M?", value=True)]
).queue().launch(server_name="0.0.0.0")